Where is AI and Data Center Spend Coming from?

AI and digital transformation are driving capital reallocation, but it is more complicated than “AI spending is taking money directly away from everything else.” The biggest source of money is Big Tech’s operating cash flow, increasingly supplemented by debt, equity issuance, leases, infrastructure partnerships, and private capital. At the same time, the areas being squeezed the most are labor, some traditional IT/software spending, lower-priority corporate projects, and shareholder free cash flow.

The scale is remarkable. Microsoft, Alphabet, Amazon, Meta, and Oracle are on a trajectory in which AI-related infrastructure investment is becoming comparable to — and in some cases larger than — the free cash flow they have historically generated after capital spending. Reuters estimates that the major hyperscalers are headed toward spending hundreds of billions of dollars annually, with roughly $725–750 billion of 2026 spending frequently cited for the largest players.

Where the money comes from

Initially, much of the buildout was essentially financed internally:

• Advertising → AI infrastructure at Google and Meta.
• AWS/cloud profits → AI infrastructure at Amazon.
• Microsoft enterprise software/cloud cash flow → AI infrastructure at Microsoft.

Those businesses throw off enormous operating cash flow, which allows AI investment to expand without much visible sacrifice elsewhere. But we are now moving into a second phase. Internal cash generation isn’t enough to fund every desired project without consequences. Debt markets are becoming much more important. U.S. corporate AI-related debt issuance has reportedly reached about $220 billion in 2026, versus just $12.5 billion in 2025. Alphabet, Amazon, and others are issuing bonds partly to preserve cash while maintaining their infrastructure programs.

Alphabet, for example, has been raising debt both internationally and domestically. Its recent Australian-dollar bond sale raised roughly US$3.9 billion, even as the company had already experienced negative quarterly free cash flow due to its enormous capital program. There is also a huge amount of off-balance-sheet financing occurring through data-center leases, long-term chip contracts, power-purchase commitments, and infrastructure joint ventures. That shifts some of the financing burden to private equity, infrastructure funds, utilities, developers, and lenders, rather than placing all the assets directly on the hyperscaler’s balance sheet.

So the financing chain increasingly looks like:

Corporate cash flow → bonds → private credit/infrastructure funds → leases → utilities → data centers → Nvidia/AMD/chip suppliers.

That is important because AI has begun to draw funding not only from corporate IT budgets but also from global capital markets.

So what is getting squeezed?

1. People are among the clearest casualties

Meta provided perhaps the clearest description of the tradeoff. Mark Zuckerberg told employees that Meta essentially has two major cost pools: compute infrastructure and people. If more capital goes into compute, less is available for the other side. Reuters reported his comments as Meta was planning layoffs amid rising AI capital spending. That isn’t theoretical. U.S. technology companies have cut roughly 140,000 jobs in 2026, even as they commit unprecedented sums to AI infrastructure. Amazon, Oracle, Meta, and Microsoft account for a significant portion of those reductions. Some cuts are post-pandemic restructuring rather than AI substitution, but capital reallocation toward AI is clearly part of the story.

This is one of the most important economic transformations happening. Technology companies are becoming more capital-intensive and relatively less labor-intensive. For decades, software was extraordinarily asset-light.

Hire engineers → write software → sell another copy at nearly zero marginal cost.

AI changes the equation:

engineers + GPUs + data centers + electricity + networking + cooling + power plants.

Big Tech is beginning to look partly like industrial infrastructure.

2. Traditional corporate IT spending is being reprioritized

There is also evidence that AI infrastructure is crowding out other IT expenditure. IBM recently warned that companies were redirecting spending toward servers, chips, networking and AI infrastructure, delaying or reducing some traditional software purchases.

That means CIOs increasingly face something resembling:

Not every category loses. In fact, data modernization, cybersecurity, and systems integration can be beneficial, as organizations find that fragmented systems and poor data hinder AI deployment. Capgemini reports that AI is dynamically accelerating the modernization of legacy applications and data architecture. So the distinction is: AI-adjacent spending is booming. IT projects that don’t contribute to AI readiness are increasingly vulnerable.

3. Free cash flow and shareholder returns are being squeezed

This may be the highest immediate financial cost. Alphabet, Meta, Microsoft, Amazon, and Oracle have historically generated enormous free cash flow. AI capex is consuming much more of it. Reuters estimates hyperscaler capital spending may increase by approximately $534 billion through 2027, while incremental operating cash flow rises only around $340 billion. That is approximately $1.57 of additional investment for every $1 of additional operating cash generation. Meta provides an extreme example. Its quarterly free cash flow recently fell by about 91% year over year, largely due to infrastructure investment, while its 2026 capex guidance was set at $130–145 billion.

That means there is less money available for buybacks, dividends, acquisitions, employee compensation, and unrelated investment. For now, several hyperscalers continue to buy back shares and pay dividends, but mathematically, something must eventually give if AI investment grows faster than cash generation.

4. Capital markets themselves are beginning to feel it

This is the less obvious effect. If Amazon or Alphabet raises tens of billions in bonds, those investors have to obtain the money somewhere. A pension fund purchasing $2 billion of Alphabet bonds does not simultaneously invest that $2 billion in industrial companies, municipal bonds, real estate, small-business lending, utilities, Treasuries, or other corporate debt. Reuters is already reporting signs of investor fatigue as AI-related bond supply grows, with widening spreads and companies having to offer better yields to attract buyers. That produces a genuine crowding-out mechanism. Not necessarily dramatic yet, but economically important.

Capital is moving toward

GPUs → servers → networking → data centers → electrical generation → transmission → cooling → AI software

And away from the marginal project somewhere else.

5. Electricity infrastructure is becoming another hidden AI expense

This is particularly interesting from an operations perspective. The AI investment isn’t confined to technology companies. Utilities must build generation, transmission, substations, transformers, gas turbines, battery storage, and grid interconnections. EIA now explicitly identifies data centers as a major driver of U.S. electricity-demand growth after roughly 15 years of relatively stagnant consumption. That means part of the AI buildout gets socialized across a much larger capital system.

Utilities borrow and invest billions because Microsoft, Meta, Google, and Amazon need electricity. And depending upon the regulatory structure, some portions ultimately reach electricity customers through rates. Thus, households and ordinary industrial manufacturers can indirectly participate in financing AI infrastructure even if they never purchase an AI product.

The bigger economic picture

I would characterize what is happening as a massive rotation in capital formation rather than simply an increase in spending.

Think about a hypothetical $100 of corporate resources five years ago:

2021

$30 people
$20 conventional IT
$15 facilities
$15 acquisitions
$10 dividends/buybacks
$10 miscellaneous investment

Today, the same company might effectively be saying:

2026

$25 people
$15 conventional IT
$35 AI/data infrastructure
$10 facilities
$10 shareholder returns
$5 miscellaneous

Those numbers are illustrative, but the directional shift is increasingly visible. The most important insight is that AI investment doesn’t have to reduce consumer spending immediately to create an opportunity cost.

The first order crowding out occurs within corporate capital allocation:

AI capex ↑
→ free cash flow ↓
→ hiring/headcount ↓
→ non-AI IT projects ↓
→ borrowing ↑
→ demand for capital ↑

Then comes the second-order effect:

data-center demand ↑
→ electricity infrastructure ↑
→ construction/material demand ↑
→ transformers/turbines/chips constrained
→ prices and interest rates potentially ↑

Only afterward does some of that reach households.

The very interesting manufacturing implication

This may become one of the defining industrial capital allocation problems of the late 2020s. A dollar invested in an AI data center competes at the margin with a dollar invested in a new machining center, warehouse automation, factory expansion, new production line, fleet replacement, housing development or conventional commercial real estate. We haven’t yet reached the point where I would say “AI is starving U.S. manufacturing of capital.” But we are clearly reaching the point where AI is competing with traditional industrial investment for electricity, transformers, construction labor, engineering talent, copper, turbines, financing, and physical infrastructure.

That is a much bigger issue than the commonly discussed question of whether AI will eliminate jobs. AI may be becoming one of the largest reallocations of productive capital since the telecom/internet buildout of the 1990s. And unlike the original software revolution, this one is extraordinarily physical. Watch manufacturing capex, non-AI IT investment, corporate headcount, free cash flow, and electricity prices over the next two years. Those are probably where the opportunity cost will become most visible.

Dr. Chris Anderson President

CEO bizmasterz.com, Lean Six Sigma Black Belt, Quality Consultant, Industrial Intelligence / Operations Management / Forecasting / QMS / Agentic AI expertise.

See Full Bio